Introduction

Fluid accumulation is common in critically ill children and is associated with increased mortality and morbidity [1,2,3,4]. It is generally calculated using fluid balance charts and patient weight, as documented in the medical record [1, 3, 4]. In contrast, edema is a clinical sign that describes excessive extravascular fluid within tissues and is defined subjectively based on the presence of skin tissue edema (limb, sacral, periorbital), pulmonary edema (crackles, increased ventilatory pressures, pleural effusions), and/or ascites. Though in practice edema and fluid accumulation (as determined by fluid balance charts) often coexist [5], it is prudent to distinguish between these terms as a positive fluid balance (“fluid accumulation”) does not necessarily imply increased excessive extravascular fluid content (“edema”) and vice versa.

In contrast to fluid accumulation, there is a paucity of literature pertaining to edema in the critical care setting. Little is known regarding the incidence, timing, relationship to chart-determined fluid accumulation, or factors associated with outcomes related to edema. Edema may be identified within the electronic medical record as part of the physical examination documented by intensive care clinicians. Therefore, natural language processing techniques could be used to better understand the epidemiology of edema. Accordingly, we aimed to describe the frequency and patterns of documentation of edema, to investigate the clinical and demographic factors associated with the documentation of edema, and to perform an exploratory analysis investigating the relationship between edema documentation and fluid balance.

Methods

Study design, setting, and population of interest

We performed a retrospective study of critically ill children (aged < 18 years) admitted for at least 12 h to a general or cardiac pediatric ICU at The Royal Children’s Hospital (RCH) in Melbourne, Australia, between 1 August 2016 and 31 July 2021. We selected a minimum duration of 12 h so that at least one documented nursing note and one documented medical note were available for each patient. For patients with multiple ICU admissions throughout the study period, individual admissions were considered unique episodes. This study was approved by the RCH Human Research Ethics Committee (RCH HREC number QA/75327/RCHM-2021, “Edema Description in Pediatric Critical Care: Terms, Patterns and Clinical Characteristics), and procedures to conduct this study were followed in accordance with the ethical standards of the RCH Human Research Ethics Committee and with the Helsinki Declaration of 1975. All reporting was performed in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [6].

Data acquisition and management

Data on patient demographics (age, sex, admission diagnosis [as categorized by Australian and New Zealand Paediatric Intensive Care Registry diagnostic codes]) and clinical characteristics were extracted from the ICU database. Clinical data included admission weight, Paediatric Index of Mortality 3 [7], Risk-Adjusted Congenital Heart Surgery (RACHS) score, mechanical ventilation (invasive and non-invasive), extracorporeal membrane oxygenation (ECMO), renal replacement therapy (RRT), length of stay (ICU, hospital), and ICU survival status (dead vs alive at ICU discharge).

Fluid balance data were collected from the electronic medical record (EMR) (EPIC, VA, USA). Daily intake and output as well as net overall daily fluid balance were extracted. Fluid accumulation was expressed as a cumulative percentage during PICU stay [1, 4]. This was calculated as ([cumulative fluid intake (L) − fluid output (L)]/PICU admission weight × 100) as previously reported [1, 8].

Typed progress notes from intensive care medical and nursing staff were extracted using SQL from the EMR reporting database Epic Clarity© hosted on Microsoft SQL Server 2019© (Microsoft Corporation, WA, USA). Each note was subject to language analysis performed using Python 3.9 (Python Software Foundation, Python Language Reference, www.python.org)). These analysis steps were as follows: (1) search each note for the identified edema primary and secondary terms (defined in the supplementary digital content); (2) for each identified edema term, extract the preceding term to assess if this was a negation term (such as “no”, “non”, or “not”); and (3) identify if any of the exclusion terms (supplementary digital content) were included in that note. This language analysis resulted in a per note list of integer variables indicating the frequency of the primary and secondary terms and the negated variates of these terms and the presence or absence of the exclusion terms. Only notes written by primary ICU providers were included. Primary ICU providers included ICU-attending providers, ICU trainees, and registered ICU nurses. Time from ICU admission to edema documentation, and frequency of documentation, were recorded. ICU provider type was also abstracted from the EMR and classified as “ICU-attending provider”, “ICU trainee, MD”, and “registered nurse”. Conventional documentation in this institution occurs a minimum of twice per 24 h by primary ICU providers resulting in documentation of patient assessment every 4–6 h.

Edema terms

Terms describing edema were selected, a priori, through literature review and expert clinical consensus (“primary edema terms”, Supplementary Table 1). Terms were then divided into primary and secondary edema terms. Primary terms refer to those terms that included the root term “edema” (and its derivatives) or terms that referenced a fluid overloaded state (Supplementary Table 1) Secondary terms included other commonly used terms and synonyms that are used to describe edema (Supplementary Table 1). Words suggestive of cerebral edema, angioedema, lymphedema, or myxedema were excluded. Terms that indicated edema resolution (e.g., “resolved” and “cleared”) or negation terms (e.g., “no”, “nil” and “not”) were included in the text search. A documented edema episode was defined as the presence of a medical record entry with at least one “primary” edema term that did not have a negation term before or after the primary edema term. Instances in which more than one edema term was found in the note were included in the edema documentation group so long as at least one “primary” edema term was included that did not have a negation term before or after the primary edema term. We then searched for secondary edema terms (Supplementary Table 1) which were then included for analysis of types and frequency of edema descriptions. Notes that did not include any of the primary edema terms or those used a negation term before or after the edema term were not included in final counts of edema description.

Outcomes

The primary outcome was the proportion of patient admissions with documented edema. Secondary outcomes included the clinical characteristics of children with documented edema, the frequency and timing of edema documentation, and the frequency of primary and secondary terms used by clinician type. In addition, clinical characteristics of children with documented edema compared to those without. Lastly, in an exploratory analysis, we investigated the relationship between edema and fluid balance in those with and without documented edema.

Analysis

Demographic and clinical characteristics were described using frequencies and percentages (categorical variables), means and standard deviations (SD; continuous and normally distributed variables), or medians and interquartile ranges (IQR; continuous and skewed variables). For the primary outcome, edema documentation was expressed a dichotomous outcome (yes vs no throughout entirety of ICU stay) and for secondary outcomes as the number of notes containing a primary edema term as a percentage of the total notes, published per patient, by ICU provider type. The Wilcoxon-signed rank test was performed to investigate the difference between fluid accumulation and edema documentation on individual days. The resulting p-values were corrected for the multiple test situations with the Bonferroni formula.

Secondary analysis of characteristics associated with edema documentation was assessed using multivariable logistic regression. Multicollinearity, defined as a tolerance value of less than 0.25, was assessed for each of the covariates, and subsequently, some covariates were excluded. Results were reported as adjusted odds ratios (aOR) and 95% confidence intervals (CI). The relationship between edema documentation and mortality was also investigated using multivariable logistic regression reporting aOR and 95% CI. Instances in which more than 1 edema term was documented in a patient note, Pearson’s correlation coefficients were calculated and reported to explore the co-documentation for each possible pair of terms. Lastly, the relationship between daily fluid balance and edema status was explored using a linear mixed-effect model that included a random intercept and random slope of ICU Day with fixed effects including edema status, ICU day, and edema-ICU Day interaction term, and only the p-value significance is reported. Data were analyzed using R software, version 4.1.0 (the R Foundation for Statistical Computing, Vienna, Austria). R packages are listed in the supplemental digital content (Supp. Table 2). Only individuals with complete data were included in the final analyses; no imputation for missing data was performed. Two-sided p-values < 0.05 were considered statistically significant.

Results

There were 8949 patient admissions between August 1, 2016, and July 31, 2021, and a total 514,482 medical record entries. A total of 7884 eligible patients were included with a total of 211,122 patient notes written by primary ICU providers (Supp. Fig. 1). The demographic and clinical characteristics of all admissions and by the presence of edema is shown in Table 1. The median (IQR) age of children studied was 678 (128–29,53) equivalent to 2 years (4 months to 8 years) with a median (IQR) length of ICU stay of 33 (12–98) h. More than one-third (39%) of patient admissions had a diagnosis pertaining to cardiovascular disease (surgical and non-surgical) on admission (Table 1). Amongst all included patients, there was a total of 46,730 ICU days studied.

Table 1 Demographic and clinical characteristics of children based on documentation of edema
Full size table

Characteristics of edema documentation

A total of 3087 (39%) patient admissions had one or more documentations of edema (Table 1) with a median (IQR) of 2 (1–4) notes per admission containing a primary edema term. The primary terms containing “edema” accounted for 96% of all identified admissions with documented edema. Volume overload and fluid overload alone were identified in approximately 1% and 3% of admissions with documented edema respectively. Eighty percent of patients with at least one documentation of edema had edema documented in less than 15% of their total notes (Supp. Fig. 2).

“Edema” was the most common primary term used by all provider types. Figure 1 shows the proportion of documented edema terms by provider type. ICU nurses documented a total of 140,990 (67%) of all included patient notes, approximately twice as frequently as ICU trainees who documented a total of 58,689 (28%) of notes and nearly twelve times more frequently than ICU attendings who documented a total of 11,443 (5%) of all notes. The most common secondary term for nursing staff and ICU trainees to describe edema was “pitting”, documented in 9% and 4% of edema descriptions. The most common secondary term to describe edema amongst ICU attendings was “swelling” documented in 4% of descriptions.

Fig. 1
figure 1

Percentage of primary and secondary terms used to describe edema by provider. The types of edema descriptions in documentation are displayed. The proportion of provider notes containing the various descriptions of edema is represented as well as the terminology used to describe edema amongst various provider types as a proportion of the total notes with documented edema by provider in patients in whom edema was documented. A total of 13,186 notes were included. Of note, some notes contained more than one term. For the purposes of this description, we illustrate the types of edema descriptions in critical care. See supplementary Fig. 5 for correlations between terms within the same notes

Full size image

The proportion of admissions with documentation of edema, per day of intensive care admission, is shown in Supp. Fig. 3. Twenty percent of edema documentation occurred on admission, and 49% occurred before day 2. In all diagnostic groups, edema documentation appeared to peak within the first 2 days of intensive care admission (Fig. 2). There was a total of 3184 (2%) of all included patient notes in which two different terms were documented. In exploratory analysis, there was negligible correlation found between various terms documented together (Supp. Fig. 5) with the largest correlation coefficient found between edematous and edema and pitting and edema both with a Pearson’s correlation coefficient = 0.25.

Fig. 2
figure 2

Time from admission to first documentation of edema for all diagnostic groups (n = 3087 admissions). Shows the timing of edema documentation per ICU Day by admission diagnosis. Day 0 represents the day of admission. Overall, the majority of patients first documentation of edema occurs on day 1. Edema documentation will also commonly occur on the day of admission (day 0) for most patients in which edema will be documented. A total of 3087 patients were found to have edema documented at least once

Full size image

Table 1 shows the differences in demographic, clinical characteristics, and cumulative fluid balance in admissions with and without documented edema. In multivariable regression analyses, factors associated with increased odds of edema documentation included aOR (CI) mechanical ventilation 2.6 (2.0–3.0), NIV 1.6 (1.3–2.0), ECMO 3.8 (2.0–6.0), RRT 4.8 (4.0–6.0), and vasoactive support 3.1 (3.0–4.0) (Fig. 3). In univariate analysis, cumulative fluid balance at 48 h was not associated with death (p = 0.349); however, in multivariable regression analyses, after adjusting for confounders, relevant covariates, and cumulative fluid balance at 48 h, edema documentation was associated with an increased odd of death: aOR 1.6 [95% CI; 1.1–2.1] (Supp. Fig. 4). Notably, respiratory diagnosis was removed from the model due to collinearity.

Fig. 3
figure 3

Multivariable logistic regression of factors associated with edema documentation. Outlines results of the multivariable regression analysis aiming to investigate the association between patient factors, intensive care therapies, and diagnoses with any documentation of edema. Diamond points represent odds ratios, and the surrounding bars are the 95% confidence intervals. PIM3, Paediatric Index of Mortality 3; MV, invasive mechanical ventilation; NIV, noninvasive ventilation; ECMO, extracorporeal membrane oxygenation; RRT, renal replacement therapy. C stat, 0.74

Full size image

Fluid balance

Median fluid balance in the first 7 days of ICU stay was not different between groups in univariate analysis (Table 1). Figure 4 shows that in those with edema, the median fluid accumulation peaked on day 2 and decreased thereafter. Between days 3 and 7, those with edema had lower fluid balance compared to those without edema. The linear mixed-effect model exploring the relationship between daily fluid balance and edema status showed that as ICU Day increased, median cumulative fluid balance decreased (p < 0.001) and, in patients with documented edema, as ICU Day increased, there was a greater decrease in cumulative daily fluid balance (p < 0.001).

Fig. 4
figure 4

Daily percentage fluid balance for the first 7 days in those with and without edema for the whole cohort. Compares the fluid balance percent (median, IQR) daily within the ICU between patients in whom edema was documented as “present” and in those in whom it was not documented or documented as “not present”. Days labelled as “*” represent statistically significant adjusted p-values using the Wilcoxon signed-rank test and the correction for the multiple test situation with the Bonferroni test. Fluid accumulation percent was calculated as follows: \(\mathrm{cumulative\, fluid\, balance\, }(\mathrm{\%}) =\frac{[\mathrm{cumulative\, daily\, fluid\, intake\, }\left(\mathrm{all\, sources}\right)\left(\mathrm{L}\right)-\mathrm{cumulative\, daily\, fluid\, output }(\mathrm{all\, sources}) (\mathrm{L})]}{\mathrm{PICU\, admission\, weight\, }(\mathrm{kg})}*100\)  

Full size image

Discussion

In this retrospective study, we analyzed over 200,000 electronic medical record entries of intensive care specialists, trainees, and nurses to describe the patterns of documentation of edema. Edema was documented in almost 40% of admission. The three main findings include the broad and subjective range of terms used by clinicians to describe the clinical state of edema, the clinical timing of edema through description, and the paradoxical relationship between documented edema and fluid balance.

First, this study demonstrates the broad and subjective range of terms used by clinicians to describe the clinical state of edema (Fig. 2). “Fluid overload”, “pitting”, and “swelling” were commonly used terms accounting for nearly a quarter of all edema descriptions by ICU physicians. The characterization, description, and utilization of the clinical state of “edema” is also scarce and varied within the literature, most commonly referring to chest wall edema or the presence of eyelid edema [9,10,11]. Without objective methods for defining and measuring edema, investigating its effect on clinical outcomes and assigning appropriate interventions will remain a challenge.

Second, we found that edema is commonly recognized early during intensive care admission, in many diagnostic categories, and may independently be associated with harm. This is in keeping with current literature both in pediatric intensive care. There is a growing understanding that fluid accumulation can have a significant impact on mortality and morbidity in the ICU [1, 4]. Children with fluid accumulation have been shown to have poorer oxygenation [12], more days of mechanical ventilation, increased risk of major kidney injury, and longer length of ICU stay [1]. Patterns of edema recognition within various diagnostic categories were varied in this study. Children admitted with respiratory, GI, and renal diagnoses were more likely to have edema documented on the day of admission as opposed to children following surgery (cardiac and non-cardiac) who were more likely to have edema documented on day 1 or 2 of admission. This could be a representation of the natural history of various pathophysiological states. Furthermore, the multivariable regression analysis demonstrated the odds of edema documentation were also increased in biological plausibility and physiologically appropriate clinical states (e.g., edema documentation in the context of RRT). Further exploration and understanding of these patterns of edema within various patient populations or the timing of edema documentation and the initiation of various extracorporeal supports could provide insight into prevention of fluid accumulation in the tissues and guide subsequent auspicious timing of removal strategies [13, 14].

Third, this study found that admissions with documented edema, compared to those without, have lower daily fluid accumulation per day from day 3 of intensive care admission onward. This may reflect the effect of early recognition and therapeutic interventions as most documentation of edema occurred within 2 days of ICU admission. Moreover, those with edema had a much higher proportion receiving RRT, implying the possible use of this therapy for fluid removal. This could also represent the distinction between charted fluid balance and tissue edema. For example, the increased fluid balance in those without documented edema may represent the disease state, the integrity of the endothelium, and appropriate intravascular volume or enteral administration of volume regulated efficiently and distinctly by the GI tract. This is further demonstrated in the results of multivariable analysis. After accounting for intensity of therapy, size, fluid accumulation, and admission diagnosis, the recognition and documentation of edema were associated with increased mortality. To date, studies have largely focused on the timing, rate of accumulation, and consequences of fluid overload and accumulation, especially in specific subpopulations of critically ill children [1,2,3]. This study begins to explore the association between edema and patient outcomes.

Strengths and limitations

The strengths of this study include that, for the first time, it broadly describes the patterns of edema in children in intensive care. We used a large dataset and specific language analysis tools and, developed, clear characterization of edema terms. We described these patterns for 3 different provider types and performed exploratory analyses regarding the associations with fluid balance and outcomes.

This study also has several limitations. First, this was a single-center, retrospective analysis. However, this is the first study to describe the recognition of edema by clinicians, yet we acknowledge its exploratory nature. Second, the primary and secondary terms were selected subjectively and were conservative. However, we based these on findings from previous literature and commonly used terms by clinicians at this center. We also used negation and exclusion terms to improve the sensitivity of the search. Third, though we were able to extract the documentation of edema, we did not search for the severity of edema, since there is currently no standardized scale or reporting measure for this. Fourth, though this is a commonly reported metric and methodology [1, 4, 15] for calculating fluid accumulation, there are inherent limitations from documented patients’ charts and information such as unmeasured losses (e.g., unmeasured urine output and insensible losses), and measures such as change in patient weight could serve as improved measurements of fluid accumulation however were not available in this dataset. Fifth, the authors acknowledge that clinically, terms used to describe edema (e.g., volume overload or fluid overload) are often used interchangeably. To mitigate this effect on the analysis, the authors limited the primary terms and elected to only incorporate those most representative of the current clinical state. As a result, we elected to include the documentation of fluid overload in the primary terms as it is commonly used by clinicians in practice.

Lastly, there could be instances in which edema was present but not documented representing human error in documentation. We hope to mitigate this with the frequency and the redundancy of documented notes by ICU providers which amounts to documentation of patient assessment every 4–6 h.

This retrospective study observed that edema is frequently recognized and documented in critically ill pediatric patients. Edema is acknowledged early in ICU admission and most commonly documented as “edema” across different ICU providers. Further prospective work on the objective assessment and uniform definition of edema, differentiation and relationship to fluid accumulation, concurrently documented terms that describe clinical edema, and variation in patterns of edema within patient subgroups could provide standardization of its measurement, quantification of response to interventions to mitigate tissue edema, and improve overall fluid stewardship for critically ill children.